DOCSLIB.ORG

Performance Evaluation of a Distributed Storage Service in Community Network Clouds

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text

Load more

CONCURRENCY AND COMPUTATION: PRACTICE AND EXPERIENCE Concurrency Computat.: Pract. Exper. (2015) Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/cpe.3658 SPECIAL ISSUE PAPER Performance evaluation of a distributed storage service in community network clouds Mennan Selimi1,*,†, Felix Freitag1, Llorenç Cerdà-Alabern1 and Luís Veiga2 1Department of Computer Architecture, Universitat Politècnica de Catalunya, Barcelona, Spain 2Instituto Superior Técnico (IST), INESC-ID Lisboa, Lisbon, Portugal SUMMARY Community networks are self-organized and decentralized communication networks built and operated by citizens, for citizens. The consolidation of today’s cloud technologies offers now, for community networks, the possibility to collectively develop community clouds, building upon user-provided networks and extend- ing toward cloud services. Cloud storage, and in particular secure and reliable cloud storage, could become a key community cloud service to enable end-user applications. In this paper, we evaluate in a real deploy- ment the performance of Tahoe least-authority file system (Tahoe-LAFS), a decentralized storage system with provider-independent security that guarantees privacy to the users. We evaluate how the Tahoe-LAFS storage system performs when it is deployed over distributed community cloud nodes in a real community network such as Guifi.net. Furthermore, we evaluate Tahoe-LAFS in the Microsoft Azure commercial cloud platform, to compare and understand the impact of homogeneous network and hardware resources on the performance of the Tahoe-LAFS. We observed that the write operation of Tahoe-LAFS resulted in similar performance when using either the community network cloud or the commercial cloud. However, the read operation achieved better performance in the Azure cloud, where the reading from multiple nodes of Tahoe- LAFS benefited from the homogeneity of the network and nodes. Our results suggest that Tahoe-LAFS can run on community network clouds with suitable performance for the needed end-user experience. Copyright © 2015 John Wiley & Sons, Ltd. Received 28 February 2015; Revised 22 June 2015; Accepted 8 August 2015 KEY WORDS: community networks; community cloud; cloud storage 1. INTRODUCTION Wireless community networks [1] encompass an emergent model of infrastructure that aims to sat- isfy a community’s demand for Internet access and information and communications technology services. Most community networks originated in rural areas, which commercial telecommuni- cation operators left behind when deploying the broadband access infrastructure for urban areas. Different stakeholders of such a geographic area teamed up to invest, create, and run a community network as an open telecommunication infrastructure based on self-service and self-management by the users. Since the first community networks started more than 10 years ago, they have become rather successful. Several large community networks in Europe have from 500 to 30,000 nodes; these include Guifi.net [2], FunkFeuer [3], Athens Wireless Metropolitan Network (AWMN) [4], and many more that exist worldwide. Most of them are based on Wi-Fi technology (ad-hoc net- works, IEEE 802.11a/b/g/n access points in the first hop, long-distance point-to-point Wi-Fi links for the trunk network), and also a growing number of optic fiber links have started to be deployed. *Correspondence to: Mennan Selimi, Department of Computer Architecture, UPC-BarcelonaTech, Jordi Girona, 1-3, 08034, Barcelona, Spain. †E-mail: [email protected] Copyright © 2015 John Wiley & Sons, Ltd. M. SELIMI ET AL. Figure 1. Guifi.net and Athens Wireless Metropolitan Network nodes and links in the area around Barcelona and Athens. Figure 1 shows as an example the wireless links and nodes of the Guifi.net and AWMN community network in the area of Barcelona and Athens. Community networks are a successful case of resource sharing among a collective, where resources shared are not only the networking hardware but also the time, effort, and knowledge contributed by its members that are required for maintaining the network. Resource sharing in com- munity networks from the equipment perspective refers, in practice, to the sharing of the nodes network bandwidth. This sharing enables the traffic from other nodes to be routed over the nodes of different node owners. This is carried out in a reciprocal manner, which allows community net- works to successfully operate as IP networks. The sharing of other computing resources such as storage, which is now common practice in today’s Internet through cloud computing, hardly exists in community networks, but it can be made possible through clouds in community networks, that is, community network clouds [5]. The concept of community network clouds has been introduced in its generic form before [e.g., 6, 7], as a cloud deployment model in which a cloud infrastructure is built and provisioned for an exclusive use by a specific community of consumers with shared concerns and interests. Besides public, private, and hybrid models of cloud computing, a community cloud presents another option differing from the others in that it is designed with a specific community in mind and with costs and responsibility shared among community members. Security concerns, government legislation, need for sophisticated control, or enhanced performance requirements in many scenarios make it difficult to rely on public clouds, and private clouds are not always the most cost-effective option. We refer here to a specific kind of a community cloud in which sharing of computing resources is from within community networks [8, 9], using the application models of cloud computing in general. Establishing a community cloud not only involves many challenges, both in technological context and socio-economic context, but also promises an interesting value proposition for communities in terms of local services and applications. To conduct our evaluation of applications in a realistic scenario, our approach is to leverage on the cloud infrastructure provided by an ongoing cloud deployment in the Guifi.net community network [10] and a testbed deployed in community networks [1, 11]. The approach for performance assessment of applications in community networks, which we propose in this paper, is to set the experimental conditions as seen from the end user: experiment in production community networks, focus on metrics that are of interest to end users, and deploy applications on real nodes integrated in community networks. We specifically look at a distributed secure storage application, Tahoe least-authority file sys- tem (Tahoe-LAFS) [12], to be used within these wireless community-owned networks. After basic connectivity, storage is the most general and fundamental resource to support a distributed system, for example, to store users’ files, as in services like Dropbox, being fundamental for cloud take-up in community network scenarios. Tahoe-LAFS is an open-source distributed cloud storage system. Features of Tahoe-LAFS that are relevant for community networks are that it encrypts data at the Copyright © 2015 John Wiley & Sons, Ltd. Concurrency Computat.: Pract. Exper. (2015) DOI: 10.1002/cpe DISTRIBUTED STORAGE SERVICE IN COMMUNITY NETWORK CLOUDS client side, erasure codes it (data are broken into fragments, expanded, and encoded with redun- dant data pieces), and then disperses it among a set of storage nodes. This approach of Tahoe-LAFS results in high availability; for example, even if some of the storage nodes are down or taken over by an attacker, the entire file system continues to function correctly while preserving privacy and secu- rity. This is key for users’ adoption. We host the Tahoe-LAFS distributed application on nodes of a community cloud spread inside the production community network. Furthermore, we compare the performance of Tahoe-LAFS when it is deployed in community clouds and in a commercial cloud such as Microsoft Azure. The contributions of this paper are the following: (1) Verify the correct operation and adequate performance of the Tahoe-LAFS distributed file system in a real cloud system in a wireless com- munity network. (2) Characterize the write and read performance of Tahoe-LAFS with different workloads generated by IOzone storage benchmark. (3) Present and show the feasibility of a case for storage sharing with performance evaluation conducted in a real setting of a deployed cloud sys- tem over a community network. We expect that these results will open the door to further application deployments and take-up of community clouds. The operational distributed file system, which we investigated in this paper, should encourage application developers to create innovative cloud-based community network services and users to contribute resources that enable the hosting and usage of these services. The rest of the paper is organized as follows. Section 2 defines the community clouds and dis- cusses the requirements for such clouds. Section 3 explains the characteristics of the Tahoe-LAFS distributed file system that we use. Section 4 explains the experiment environment for community clouds and Azure cloud. Section 5 shows and discusses the results from our cloud deployment. Section 6 talks about related work, and Section 7 concludes and discusses future research directions. 2. CLOUDS IN COMMUNITY NETWORKS 2.1. Community cloud scenarios Our proposition is to deploy a distributed storage service in community networks using cloud- based resources. In this section, we
Recommended publications
  • Globalfs: a Strongly Consistent Multi-Site File System

    Globalfs: a Strongly Consistent Multi-Site File System

    GlobalFS: A Strongly Consistent Multi-Site File System Leandro Pacheco Raluca Halalai Valerio Schiavoni University of Lugano University of Neuchatelˆ University of Neuchatelˆ Fernando Pedone Etienne Riviere` Pascal Felber University of Lugano University of Neuchatelˆ University of Neuchatelˆ Abstract consistency, availability, and tolerance to partitions. Our goal is to ensure strongly consistent file system operations This paper introduces GlobalFS, a POSIX-compliant despite node failures, at the price of possibly reduced geographically distributed file system. GlobalFS builds availability in the event of a network partition. Weak on two fundamental building blocks, an atomic multicast consistency is suitable for domain-specific applications group communication abstraction and multiple instances of where programmers can anticipate and provide resolution a single-site data store. We define four execution modes and methods for conflicts, or work with last-writer-wins show how all file system operations can be implemented resolution methods. Our rationale is that for general-purpose with these modes while ensuring strong consistency and services such as a file system, strong consistency is more tolerating failures. We describe the GlobalFS prototype in appropriate as it is both more intuitive for the users and detail and report on an extensive performance assessment. does not require human intervention in case of conflicts. We have deployed GlobalFS across all EC2 regions and Strong consistency requires ordering commands across show that the system scales geographically, providing replicas, which needs coordination among nodes at performance comparable to other state-of-the-art distributed geographically distributed sites (i.e., regions). Designing file systems for local commands and allowing for strongly strongly consistent distributed systems that provide good consistent operations over the whole system.
  • Big Data Storage Workload Characterization, Modeling and Synthetic Generation

    Big Data Storage Workload Characterization, Modeling and Synthetic Generation

    BIG DATA STORAGE WORKLOAD CHARACTERIZATION, MODELING AND SYNTHETIC GENERATION BY CRISTINA LUCIA ABAD DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Computer Science in the Graduate College of the University of Illinois at Urbana-Champaign, 2014 Urbana, Illinois Doctoral Committee: Professor Roy H. Campbell, Chair Professor Klara Nahrstedt Associate Professor Indranil Gupta Assistant Professor Yi Lu Dr. Ludmila Cherkasova, HP Labs Abstract A huge increase in data storage and processing requirements has lead to Big Data, for which next generation storage systems are being designed and implemented. As Big Data stresses the storage layer in new ways, a better understanding of these workloads and the availability of flexible workload generators are increas- ingly important to facilitate the proper design and performance tuning of storage subsystems like data replication, metadata management, and caching. Our hypothesis is that the autonomic modeling of Big Data storage system workloads through a combination of measurement, and statistical and machine learning techniques is feasible, novel, and useful. We consider the case of one common type of Big Data storage cluster: A cluster dedicated to supporting a mix of MapReduce jobs. We analyze 6-month traces from two large clusters at Yahoo and identify interesting properties of the workloads. We present a novel model for capturing popularity and short-term temporal correlations in object re- quest streams, and show how unsupervised statistical clustering can be used to enable autonomic type-aware workload generation that is suitable for emerging workloads. We extend this model to include other relevant properties of stor- age systems (file creation and deletion, pre-existing namespaces and hierarchical namespaces) and use the extended model to implement MimesisBench, a realistic namespace metadata benchmark for next-generation storage systems.
  • File System Design Approaches

    File System Design Approaches

    File System Design Approaches Dr. Brijender Kahanwal Department of Computer Science & Engineering Galaxy Global Group of Institutions Dinarpur, Ambala, Haryana, INDIA [email protected] Abstract—In this article, the file system development design The experience with file system development is limited approaches are discussed. The selection of the file system so the research served to identify the different techniques design approach is done according to the needs of the that can be used. The variety of file systems encountered developers what are the needed requirements and show what an active area of research file system specifications for the new design. It allowed us to identify development is. The file systems researched fell into one of where our proposal fitted in with relation to current and past file system development. Our experience with file system the following four categories: development is limited so the research served to identify the 1. The file system is developed in user space and runs as a different techniques that can be used. The variety of file user process. systems encountered show what an active area of research file 2. The file system is developed in the user space using system development is. The file systems may be from one of the FUSE (File system in USEr space) kernel module and two fundamental categories. In one category, the file system is runs as a user process. developed in user space and runs as a user process. Another 3. The file system is developed in the kernel and runs as a file system may be developed in the kernel space and runs as a privileged process.
  • A Decentralized Cloud Storage Network Framework

    A Decentralized Cloud Storage Network Framework

    Storj: A Decentralized Cloud Storage Network Framework Storj Labs, Inc. October 30, 2018 v3.0 https://github.com/storj/whitepaper 2 Copyright © 2018 Storj Labs, Inc. and Subsidiaries This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 license (CC BY-SA 3.0). All product names, logos, and brands used or cited in this document are property of their respective own- ers. All company, product, and service names used herein are for identification purposes only. Use of these names, logos, and brands does not imply endorsement. Contents 0.1 Abstract 6 0.2 Contributors 6 1 Introduction ...................................................7 2 Storj design constraints .......................................9 2.1 Security and privacy 9 2.2 Decentralization 9 2.3 Marketplace and economics 10 2.4 Amazon S3 compatibility 12 2.5 Durability, device failure, and churn 12 2.6 Latency 13 2.7 Bandwidth 14 2.8 Object size 15 2.9 Byzantine fault tolerance 15 2.10 Coordination avoidance 16 3 Framework ................................................... 18 3.1 Framework overview 18 3.2 Storage nodes 19 3.3 Peer-to-peer communication and discovery 19 3.4 Redundancy 19 3.5 Metadata 23 3.6 Encryption 24 3.7 Audits and reputation 25 3.8 Data repair 25 3.9 Payments 26 4 4 Concrete implementation .................................... 27 4.1 Definitions 27 4.2 Peer classes 30 4.3 Storage node 31 4.4 Node identity 32 4.5 Peer-to-peer communication 33 4.6 Node discovery 33 4.7 Redundancy 35 4.8 Structured file storage 36 4.9 Metadata 39 4.10 Satellite 41 4.11 Encryption 42 4.12 Authorization 43 4.13 Audits 44 4.14 Data repair 45 4.15 Storage node reputation 47 4.16 Payments 49 4.17 Bandwidth allocation 50 4.18 Satellite reputation 53 4.19 Garbage collection 53 4.20 Uplink 54 4.21 Quality control and branding 55 5 Walkthroughs ...............................................
  • Alpha Release of the Data Service

    Alpha Release of the Data Service

    This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 777533. PROviding Computing solutions for ExaScale ChallengeS D5.2 Alpha release of the Data service Start / 01 November 2017 Project: PROCESS H2020 – 777533 Duration: 36 Months Dissemination1: Public Nature2: R Due Date: 31 January 2019 Work Package: WP 5 Filename3 PROCESS_D5.2_Alpha_release_of_the_Data_service_v1.0.docx ABSTRACT During the first 15 months of its implementation, PROCESS has progressed from architecture design based on use cases’ requirements (D4.1) and through architecture validation again based on use cases (D4.2) towards initial implementations of computing services (D6.1) and data services - effort presented here in the deliverable D5.2. In D5.2 we provide an initial demonstrator of the data services, which works in cooperation with the computation services demonstrated in D6.1. The demonstrator is based on the design of the PROCESS data infrastructure described in D5.1. The implementation and initial integration of the infrastructure are based on use case requirements, formulated here as custom application-specific services which are part of the infrastructure. The central, connecting component of the data infrastructure is LOBCDER. It implements a micro-infrastructure of data services, based on dynamically provisioned Docker containers. Additionally to LOBCDER and use case-specific services, the data infrastructure contains generic data and metadata- handling services (DISPEL, DataNet). Finally, Cloudify integrates the micro-infrastructure and the orchestration components of WP7. 1 PU = Public; CO = Confidential, only for members of the Consortium (including the EC services). 2 R = Report; R+O = Report plus Other.
  • Hands-On Linux Administration on Azure

    Hands-On Linux Administration on Azure

    Hands-On Linux Administration on Azure Explore the essential Linux administration skills you need to deploy and manage Azure-based workloads Frederik Vos BIRMINGHAM - MUMBAI Hands-On Linux Administration on Azure Copyright © 2018 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. Commissioning Editor: Vijin Boricha Acquisition Editor: Rahul Nair Content Development Editor: Nithin George Varghese Technical Editor: Komal Karne Copy Editor: Safis Editing Project Coordinator: Drashti Panchal Proofreader: Safis Editing Indexer: Mariammal Chettiyar Graphics: Tom Scaria Production Coordinator: Deepika Naik First published: August 2018 Production reference: 1310818 Published by Packt Publishing Ltd. Livery Place 35 Livery Street Birmingham B3 2PB, UK. ISBN 978-1-78913-096-6 www.packtpub.com mapt.io Mapt is an online digital library that gives you full access to over 5,000 books and videos, as well as industry leading tools to help you plan your personal development and advance your career.
  • Tahoe-LAFS Documentation Release 1.X

    Tahoe-LAFS Documentation Release 1.X

    Tahoe-LAFS Documentation Release 1.x The Tahoe-LAFS Developers January 19, 2017 Contents 1 Welcome to Tahoe-LAFS! 3 1.1 What is Tahoe-LAFS?..........................................3 1.2 What is “provider-independent security”?................................3 1.3 Access Control..............................................4 1.4 Get Started................................................4 1.5 License..................................................4 2 Installing Tahoe-LAFS 5 2.1 First: In Case Of Trouble.........................................5 2.2 Pre-Packaged Versions..........................................5 2.3 Preliminaries...............................................5 2.4 Install the Latest Tahoe-LAFS Release.................................6 2.5 Running the tahoe executable.....................................8 2.6 Running the Self-Tests..........................................8 2.7 Common Problems............................................9 2.8 Using Tahoe-LAFS............................................9 3 How To Run Tahoe-LAFS 11 3.1 Introduction............................................... 11 3.2 Do Stuff With It............................................. 12 3.3 Socialize................................................. 13 3.4 Complain................................................. 13 4 Configuring a Tahoe-LAFS node 15 4.1 Node Types................................................ 16 4.2 Overall Node Configuration....................................... 16 4.3 Connection Management........................................
  • Porting FUSE to L4re

    Porting FUSE to L4re

    Großer Beleg Porting FUSE to L4Re Florian Pester 23. Mai 2013 Technische Universit¨at Dresden Fakult¨at Informatik Institut fur¨ Systemarchitektur Professur Betriebssysteme Betreuender Hochschullehrer: Prof. Dr. rer. nat. Hermann H¨artig Betreuender Mitarbeiter: Dipl.-Inf. Carsten Weinhold Erkl¨arung Hiermit erkl¨are ich, dass ich diese Arbeit selbstst¨andig erstellt und keine anderen als die angegebenen Hilfsmittel benutzt habe. Declaration I hereby declare that this thesis is a work of my own, and that only cited sources have been used. Dresden, den 23. Mai 2013 Florian Pester Contents 1. Introduction 1 2. State of the Art 3 2.1. FUSE on Linux . .3 2.1.1. FUSE Internal Communication . .4 2.2. The L4Re Virtual File System . .5 2.3. Libfs . .5 2.4. Communication and Access Control in L4Re . .6 2.5. Related Work . .6 2.5.1. FUSE . .7 2.5.2. Pass-to-Userspace Framework Filesystem . .7 3. Design 9 3.1. FUSE Server parts . 11 4. Implementation 13 4.1. Example Request handling . 13 4.2. FUSE Server . 14 4.2.1. LibfsServer . 14 4.2.2. Translator . 14 4.2.3. Requests . 15 4.2.4. RequestProvider . 15 4.2.5. Node Caching . 15 4.3. Changes to the FUSE library . 16 4.4. Libfs . 16 4.5. Block Device Server . 17 4.6. File systems . 17 5. Evaluation 19 6. Conclusion and Further Work 25 A. FUSE operations 27 B. FUSE library changes 35 C. Glossary 37 V List of Figures 2.1. The architecture of FUSE on Linux . .3 2.2. The architecture of libfs .
  • Paratrac: a Fine-Grained Profiler for Data-Intensive Workflows

    Paratrac: a Fine-Grained Profiler for Data-Intensive Workflows

    ParaTrac: A Fine-Grained Profiler for Data-Intensive Workflows Nan Dun Kenjiro Taura Akinori Yonezawa Department of Computer Department of Information and Department of Computer Science Communication Engineering Science The University of Tokyo The University of Tokyo The University of Tokyo 7-3-1 Hongo, Bunkyo-Ku 7-3-1 Hongo, Bunkyo-Ku 7-3-1 Hongo, Bunkyo-Ku Tokyo, 113-5686 Japan Tokyo, 113-5686 Japan Tokyo, 113-5686 Japan [email protected] [email protected] [email protected] tokyo.ac.jp tokyo.ac.jp tokyo.ac.jp ABSTRACT 1. INTRODUCTION The realistic characteristics of data-intensive workflows are With the advance of high performance distributed com- critical to optimal workflow orchestration and profiling is puting, users are able to execute various data-intensive an effective approach to investigate the behaviors of such applications by harnessing massive computing resources [1]. complex applications. ParaTrac is a fine-grained profiler Though workflow management systems have been developed for data-intensive workflows by using user-level file system to alleviate the difficulties of planning, scheduling, and exe- and process tracing techniques. First, ParaTrac enables cuting complex workflows in distributed environments [2{5], users to quickly understand the I/O characteristics of from optimal workflow management still remains a challenge entire application to specific processes or files by examining because of the complexity of applications. Therefore, one low-level I/O profiles. Second, ParaTrac automatically of important and practical demands is to understand and exploits fine-grained data-processes interactions in workflow characterize the data-intensive applications to help workflow to help users intuitively and quantitatively investigate management systems (WMS) refine their orchestration for realistic execution of data-intensive workflows.
  • The Quantcast File System

    The Quantcast File System

    The Quantcast File System Michael Ovsiannikov Silvius Rus Damian Reeves Quantcast Quantcast Google movsiannikov@quantcast [email protected] [email protected] Paul Sutter Sriram Rao Jim Kelly Quantcast Microsoft Quantcast [email protected] [email protected] [email protected] ABSTRACT chine writing it, another on the same rack, and a third on a The Quantcast File System (QFS) is an efficient alterna- distant rack. Thus HDFS is network efficient but not par- tive to the Hadoop Distributed File System (HDFS). QFS is ticularly storage efficient, since to store a petabyte of data, written in C++, is plugin compatible with Hadoop MapRe- it uses three petabytes of raw storage. At today’s cost of duce, and offers several efficiency improvements relative to $40,000 per PB that means $120,000 in disk alone, plus ad- HDFS: 50% disk space savings through erasure coding in- ditional costs for servers, racks, switches, power, cooling, stead of replication, a resulting doubling of write through- and so on. Over three years, operational costs bring the put, a faster name node, support for faster sorting and log- cost close to $1 million. For reference, Amazon currently ging through a concurrent append feature, a native com- charges $2.3 million to store 1 PB for three years. mand line client much faster than hadoop fs, and global Hardware evolution since then has opened new optimiza- feedback-directed I/O device management. As QFS works tion possibilities. 10 Gbps networks are now commonplace, out of the box with Hadoop, migrating data from HDFS and cluster racks can be much chattier.
  • Extendable Storage Framework for Reliable Clustered Storage Systems by Sumit Narayan B.E., University of Madras, 2002 M.S., University of Connecticut, 2004

    Extendable Storage Framework for Reliable Clustered Storage Systems by Sumit Narayan B.E., University of Madras, 2002 M.S., University of Connecticut, 2004

    Extendable Storage Framework for Reliable Clustered Storage Systems Sumit Narayan, Ph.D. University of Connecticut, 2010 The total amount of information stored on disks has increased tremendously in re­ cent years with data storage, sharing and backup becoming more important than ever. The demand for storage has not only changed in size, but also in speed, reli­ ability and security. These requirements not only create a big challenge for storage administrators who must decide on several aspects of storage policy with respect to provisioning backups, retention, redundancy, security, performance, etc. but also for storage system architects who must aim for a one system fits all design. Storage poli­ cies like backup and security are typically set by system administrators for an entire file system, logical volume or storage pool. However, this granularity is too large and can sacrifice storage efficiency and performance - particularly since different files have different storage requirements. In the same context, clustered storage systems that are typically used for data storage or as file servers, provide very high performance and maximum scalability by striping data across multiple nodes. However, high num­ ber of storage nodes in such large systems also raises concerns for reliability in terms of loss of data due to failed nodes, or corrupt blocks on disk drives. Redundancy techniques similar to RAID across these nodes are not common because of the high overhead incurred owing to the parity calculations for all the files present on the file system. In the same way, data integrity checks are often omitted or disabled in file systems to guarantee high throughput from the storage system.
  • Collocated Data Deduplication for Virtual Machine Backup in the Cloud

    Collocated Data Deduplication for Virtual Machine Backup in the Cloud

    UNIVERSITY OF CALIFORNIA Santa Barbara Collocated Data Deduplication for Virtual Machine Backup in the Cloud A Dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Computer Science by Wei Zhang Committee in Charge: Professor Tao Yang, Chair Professor Jianwen Su Professor Rich Wolski September 2014 The Dissertation of Wei Zhang is approved: Professor Jianwen Su Professor Rich Wolski Professor Tao Yang, Committee Chairperson January 2014 Collocated Data Deduplication for Virtual Machine Backup in the Cloud Copyright c 2014 by Wei Zhang iii To my family. iv Acknowledgements First of all, I want to take this opportunity to thank my advisor Professor Tao Yang for his invaluable advice and instructions to me on selecting interesting and challenging research topics, identifying specific problems for each research phase, as well as prepar- ing for my future career. I would also like to thank my committee members, Professor Rich Wolski and Professor Jianwen Su, for their pertinent and helpful instructions on my research work. I also want to thank members and ex-members of my research group, Michael Agun, Gautham Narayanasamy, Prakash Chandrasekaran, Xiaofei Du, and Hong Tang, for their incessant support in the forms of technical discussion, system co-development, cluster administration, and other research activities. I owe my deepest gratitude to my parents, for their love and support, which give me the confidence and energy to overcome all past, current, and future difficulties. Finally and most importantly, I would like to express my gratitude beyond words to my wife, Jiayin, who has been encouraging and inspiring me since we met in love.